I’m the founder of Auriko. We ran this study to measure how much cache-aware llm routing can reduce inference costs.

Comparator names are anonymized because the point is to demonstrate the cost reduction, not to rank specific inference routers or providers.

Critiques are welcome

reply